Intelligent tutoring systems (ITS) are a successful application of cognitive science theory to the field of education. Data generated by students using an ITS can also be used to test the external validity of cognitive science principles developed largely in laboratory settings. The present paper collected data from high-school students using two versions of Cognitive Tutor, an ITS for Geometry, to assess the impact of eliminating the split-attention effect. The two versions differed in the extent to which the interface required split attention during problem solving. One version used integrated diagrams whereas the other used non-integrated tables and diagrams. Results suggested that students needed fewer problems to master skills in the integrated version, and this was particularly true for mastering difficult skills. This study demonstrates the successful use of cognitive science principles to improve learning through empirically and theoretically derived enhancements to an ITS used in a natural educational setting.